At sahibine göre kişner

The same model with the same task produces divergent outputs depending on who wrote the system prompt, what tone they encoded, and what implicit expectations they baked in. "Ownership" is the full prompt context, not just the model weights. Track prompt provenance like code authorship; behavior is not reproducible without it.